arXiv subjects

Bochao Mao

Publications and source records attributed to Bochao Mao.

3 recordsLinked to original sources

arXiv2026

On the Design of Qwen3.8-Next Architecture: Evaluation, Efficiency, and Training Stability

We describe the architecture and ablations of Qwen3.8-Flash-Next, a sparse mixture-of-experts model with 125B parameters, 6B activated per token, and additional 51B parameters of n-gram embedding tables held off the accelerator. On fourteen pre-training benchmarks the model leads the 397B-A17B predecessor on eight and trails it on the rest by at most 2.6 points, at 1/3 the activated parameters, 1/3 the training tokens, and roughly 1/9 the training FLOPs. Token mixing uses a layer-wise hybrid of Gated DeltaNet (GDN) and global attention, with one full-attention layer in every four; at continued-pretraining time those full-attention layers are replaced by Qwen Sparse Attention (QSA), which scores context at micro-block granularity with a compressed lightweight indexer. The residual stream is widened to four branches and read through an elementwise gate, a design we call the Gated Residual (GR). Capacity is added outside the backbone by a single n-gram embedding layer whose tables are prefetched from host memory. We evaluate every candidate change along three axes: loss together with downstream benchmarks; the cost of the change in training, prefill and decode; and its effect on the optimal hyperparameters and training stability. Loss and downstream accuracy do not always move together: enlarging the n-gram vocabulary lowers loss monotonically while downstream accuracy saturates. The architecture and the Muon optimizer together shift the optimal learning rate and batch size upwards, render batch-size warmup unnecessary, and substantially improve stability under stress tests. Loss, benchmarks, efficiency and stability form one design problem. Solved jointly, they yield a recipe that is simultaneously more efficient, more capable and more stable.

cs.CL↗

arXiv2024

Motif-driven Subgraph Structure Learning for Graph Classification

To mitigate the suboptimal nature of graph structure, Graph Structure Learning (GSL) has emerged as a promising approach to improve graph structure and boost performance in downstream tasks. Despite the proposal of numerous GSL methods, the progresses in this field mostly concentrated on node-level tasks, while graph-level tasks (e.g., graph classification) remain largely unexplored. Notably, applying node-level GSL to graph classification is non-trivial due to the lack of find-grained guidance for intricate structure learning. Inspired by the vital role of subgraph in graph classification, in this paper we explore the potential of subgraph structure learning for graph classification by tackling the challenges of key subgraph selection and structure optimization. We propose a novel Motif-driven Subgraph Structure Learning method for Graph Classification (MOSGSL). Specifically, MOSGSL incorporates a subgraph structure learning module which can adaptively select important subgraphs. A motif-driven structure guidance module is further introduced to capture key subgraph-level structural patterns (motifs) and facilitate personalized structure learning. Extensive experiments demonstrate a significant and consistent improvement over baselines, as well as its flexibility and generalizability for various backbones and learning procedures.

cs.LG↗

arXiv2023

OpenGSL: A Comprehensive Benchmark for Graph Structure Learning

Graph Neural Networks (GNNs) have emerged as the de facto standard for representation learning on graphs, owing to their ability to effectively integrate graph topology and node attributes. However, the inherent suboptimal nature of node connections, resulting from the complex and contingent formation process of graphs, presents significant challenges in modeling them effectively. To tackle this issue, Graph Structure Learning (GSL), a family of data-centric learning approaches, has garnered substantial attention in recent years. The core concept behind GSL is to jointly optimize the graph structure and the corresponding GNN models. Despite the proposal of numerous GSL methods, the progress in this field remains unclear due to inconsistent experimental protocols, including variations in datasets, data processing techniques, and splitting strategies. In this paper, we introduce OpenGSL, the first comprehensive benchmark for GSL, aimed at addressing this gap. OpenGSL enables a fair comparison among state-of-the-art GSL methods by evaluating them across various popular datasets using uniform data processing and splitting strategies. Through extensive experiments, we observe that existing GSL methods do not consistently outperform vanilla GNN counterparts. We also find that there is no significant correlation between the homophily of the learned structure and task performance, challenging the common belief. Moreover, we observe that the learned graph structure demonstrates a strong generalization ability across different GNN models, despite the high computational and space consumption. We hope that our open-sourced library will facilitate rapid and equitable evaluation and inspire further innovative research in this field. The code of the benchmark can be found in https://github.com/OpenGSL/OpenGSL.

cs.LG↗